DDoS Detection Method Based on Instance Transfer Learning

As a significant threat to network security, Distributed Denial of Service Attack (DDoS) is relatively simple to launch and difficult to prevent. Over the year, machine learning-based methods are widely used for DDoS detection. While, along with the evolution of attacking skill, machine learning-based detection technology’s general applicability faces severe challenges. Transfer learning is a machine learning method that seeks to overcome isolated learning tasks and use knowledge learned by one job to solve related ones. In this paper, we propose an instance-based transfer learning algorithm for DDoS attack detection. The proposed method treats publicly accessible DDoS datasets as the source domain and selects top m information gain instances for dimension reconstruction. The reconstructed selected features and target are integrated as input of the TrAdaBoost algorithm through dimension reconstruction. It provides a DDoS detection method with good general applicability for unseen attack categories detection.